ENHANCING ROAD SAFETY THROUGH ADVANCED MACHINE LEARNING: A COMPARATIVE STUDY OF RANDOM FOREST AND NEURAL NETWORKS FOR MULTI-CLASS ACCIDENT SEVERITY PREDICTION

Abstract

Traffic accidents remain a major challenge for urban planning and transportation engineering, posing continuous risks to human life and infrastructure. Despite recent advances in traffic safety measures, accurately predicting and classifying accident severity particularly under conditions of class imbalance remains difficult. This study proposes an integrated framework combining Random Forest (RF) and Artificial Neural Network (ANN) models to predict accident severity as slight, serious, or fatal. Using a large-scale UK government dataset comprising 2,047,256 accident records with 34 attributes, extensive data preprocessing was conducted, including temporal transformation, missing value imputation, and outlier removal. To mitigate class imbalance, weighted learning strategies were applied. Model performance was evaluated using confusion matrices, precision, recall, and F1-score. Results showed that while ANN and RF achieved overall accuracies of 85.99% and 70.97%, respectively, both models struggled to predict severe accidents. RF outperformed ANN in predicting fatal accidents (F1-score of 4.57% versus 0%), whereas both models performed well for slight accidents. These findings highlight the need for improved methodologies to address class imbalance in road safety severity prediction.